Semi-Supervised Crowd Counting from Unlabeled Data

TL;DR

This work tackles the expensive labeling bottleneck in image-based crowd counting by introducing $S^{4}$Crowd, a semi-supervised framework that combines a supervised density-estimation path with an unsupervised pathway comprising Crowd Scale Equivariance (CSE) and Crowd Entropy Consistency (CEC) regularizers and a crowd-driven Gated-Crowd-Recurrent-Unit (GCRU) for pseudo-label generation. The model uses a dynamic joint loss that blends $\mathcal{L}_{S}$ with unsupervised terms $\mathcal{L}_{CSE}$, $\mathcal{L}_{CEC}$, and $\mathcal{L}_{PS}$, with a schedule that increases the unsupervised contribution as pseudo-label quality improves. Extensive experiments on ShanghaiTech A/B, UCF_QNRF, and WE demonstrate that $S^{4}$Crowd consistently outperforms Gaussian-Process based and surrogate-task semi-supervised methods, with notable gains on challenging datasets and under low-label regimes. The proposed approach reduces annotation requirements while delivering robust crowd-density estimation under variations in scale and illumination, offering practical benefits for smart-city transportation analytics. Future work includes exploring domain adaptation to bridge gaps when incorporating external unlabeled data from different sources.

Abstract

Automatic Crowd behavior analysis can be applied to effectively help the daily transportation statistics and planning, which helps the smart city construction. As one of the most important keys, crowd counting has drawn increasing attention. Recent works achieved promising performance but relied on the supervised paradigm with expensive crowd annotations. To alleviate the annotation cost in real-world transportation scenarios, in this work we proposed a semi-supervised learning framework $S^{4}\textit{Crowd}$, which can leverage both unlabeled/labeled data for robust crowd counting. In the unsupervised pathway, two \textit{self-supervised losses} were proposed to simulate the crowd variations such as scale, illumination, based on which supervised information pseudo labels were generated and gradually refined. We also proposed a crowd-driven recurrent unit \textit{Gated-Crowd-Recurrent-Unit (GCRU)}, which can preserve discriminant crowd information by extracting second-order statistics, yielding pseudo labels with improved quality. A joint loss including both unsupervised/supervised information was proposed, and a dynamic weighting strategy was employed to balance the importance of the unsupervised loss and supervised loss at different training stages. We conducted extensive experiments on four popular crowd counting datasets in semi-supervised settings. Experimental results supported the effectiveness of each proposed component in our $S^{4}$Crowd framework. Our method achieved competitive performance in semi-supervised learning approaches on these crowd counting datasets.

Figures (3)

• Figure 1: The Proposed $S^{4}$Crowd Framework. It aims to provide reliable pseudo labels for unlabeled data utilizing series output, which is used to train the model along with the labeled data.
• Figure 2: The overall architecture of the proposed Unsupervised pathway, which consists of CSE/CEC regularization terms and pseudo labels generation with Gated-Crowd-Recurrent-Unit(GCRU).
• Figure 3: Gated-Crowd-Recurrent-Unit